Hungarian pseudo-hierarchies for coarse-to-fine ultra-fine-grained recognition: SoyLocal, cotton and soybean checkpoints.
AI & ML interests
None defined yet.
Recent Activity
View all activity
Token reduction with cross-layer cache aggregation for ultra-fine-grained image recognition: paper and fine-tuned ViT checkpoints.
249 checkpoints of students trained from scratch with a CAL teacher guiding data augmentation. Load with fgir-zoo.
689 parameter-efficient ViT checkpoints for ultra-fine-grained recognition: ILA, SAW and AAA adapters. Load with fgir-zoo.
SLAM-ASR models for twelve languages: a linear projector trained between a frozen Whisper-medium encoder and a frozen Tiny Aya or Qwen3-4B decoder.
-
ERISLab/LisTAya
Automatic Speech Recognition • Updated -
ERISLab/q2a_openai_whisper-medium_CohereLabs_tiny-aya-global_ws-am_et-250
4B • Updated • 35 -
ERISLab/q2a_openai_whisper-medium_CohereLabs_tiny-aya-global_ws-am_et-1000
4B • Updated • 5 -
ERISLab/q2a_openai_whisper-medium_CohereLabs_tiny-aya-global_ws-ha_ng-900
4B • Updated • 49
Backbones and pretraining for hierarchical fine-grained recognition: fine-tuned and frozen checkpoints on CUB, Aircraft, Cars.
2,083 teacher and distilled-student checkpoints for teacher selection in FGIR. Load with fgir-zoo.
242 fully fine-tuned ViT-B/16 checkpoints: FGIR heads on 21 datasets at 224 and 448 px. Load with fgir-zoo.
825 fine-grained image recognition checkpoints: 16 backbones x FT/FZ/CAL/CALMix on 17 datasets. Load with fgir-zoo.
Hungarian pseudo-hierarchies for coarse-to-fine ultra-fine-grained recognition: SoyLocal, cotton and soybean checkpoints.
Backbones and pretraining for hierarchical fine-grained recognition: fine-tuned and frozen checkpoints on CUB, Aircraft, Cars.
Token reduction with cross-layer cache aggregation for ultra-fine-grained image recognition: paper and fine-tuned ViT checkpoints.
2,083 teacher and distilled-student checkpoints for teacher selection in FGIR. Load with fgir-zoo.
249 checkpoints of students trained from scratch with a CAL teacher guiding data augmentation. Load with fgir-zoo.
242 fully fine-tuned ViT-B/16 checkpoints: FGIR heads on 21 datasets at 224 and 448 px. Load with fgir-zoo.
689 parameter-efficient ViT checkpoints for ultra-fine-grained recognition: ILA, SAW and AAA adapters. Load with fgir-zoo.
825 fine-grained image recognition checkpoints: 16 backbones x FT/FZ/CAL/CALMix on 17 datasets. Load with fgir-zoo.
SLAM-ASR models for twelve languages: a linear projector trained between a frozen Whisper-medium encoder and a frozen Tiny Aya or Qwen3-4B decoder.
-
ERISLab/LisTAya
Automatic Speech Recognition • Updated -
ERISLab/q2a_openai_whisper-medium_CohereLabs_tiny-aya-global_ws-am_et-250
4B • Updated • 35 -
ERISLab/q2a_openai_whisper-medium_CohereLabs_tiny-aya-global_ws-am_et-1000
4B • Updated • 5 -
ERISLab/q2a_openai_whisper-medium_CohereLabs_tiny-aya-global_ws-ha_ng-900
4B • Updated • 49